UCB — DT Commercial AI & Agentic Lead · Cheat Sheet

Q1 opens with commercial + a dollar figure · numbers slow, with a pause · end on the closer, then stop · full prep: UCB_Interview_Prep_Q1_to_Q5.md

Q1 — Commercial experience~35 sec

"Where is your commercial pharma experience?"

Commercial P&L first — $23M across seven states at Syncor, 15% margins, client-facing business development into hospitals and physician practices — plus business-requirements ownership for the $13M CRM running order fulfillment. AVOXI: omnichannel journeys on Salesforce Marketing Cloud, +6% retention. Street Grace: chatbots on the platform. Owned CRM requirements, built the journey layer. Pharmacist since 1984 — I've been the HCP on the receiving end of pharma engagement ever since.

1984–2005Syncor → Cardinal$23M P&L · 7 states · $13M CRM reqs · $300M claims recovery (BI at Cardinal)
2015–2017AVOXIMarketing Cloud omnichannel · +6% retention · PCI-compliant automation
2017–2018Street GraceGracie chatbot · Twilio SMS · law-enforcement handoff
2018–2021Invistics$2.1M NIH SBIR · 96% accuracy · 300+ hospitals · diversion found 160 days sooner
2025Evernorth147 CMS reports standardized · iTournament finalist, top 8 of 220

"What have you owned since Syncor?" The bar above — five chapters, unbroken record. "Just a pharmacist?" I co-authored Florida's nuclear pharmacist licensure — became the national model. I know the HCP side from the inside.

Q2 — CRM evolution · 90 days~40 sec

"What does CRM evolution mean in 2026? First 90 days?"

Two things at once: the CRM gets rebuilt on Veeva Vault CRM — end-of-support December 31, 2029, accelerated from 2030 — and its role changes: from a database reps type into, to the system of record that AI agents orchestrate around, people still approving what goes out. I've done both halves: governed GenAI layer at Amgen, compliant migration at J&J. Expansion is earned at a production milestone — the Amgen pattern.

Days 1–30Audit workflows + stack · integration points with DT AI's platform · highest-friction Patient Impact work — call prep, call notes, content approvals
Days 31–60One governed pilot — voice-to-CRM capture first · human-in-the-loop · MLR boundaries day one · baseline measured before start
Days 61–90Readout: pilot results vs. baseline · build-versus-extend-versus-buy framework · sequenced Vault CRM roadmap tied to launch calendar

UCB's head of digital technology, June 2026: "our strategic partnership with Veeva positions us well for what's next." · Veeva agent waves: Apr 2026 Safety/Quality · Aug Clinical/Reg/Medical · Dec Clinical Data.

"Never run a Vault migration?" Done the two hard parts separately — governed agent layer (Amgen), compliant multi-division migration (J&J). Vault is the combination.

Q3 — Agentic omnichannel design~32 sec

"Design an agentic solution for omnichannel HCP engagement."

The CRM stays the system of record; agents orchestrate around it. Nothing reaches an HCP without the MLR gate. Agents recommend, people approve — human-in-the-loop is the architecture; everything logged for audit. That loop is what I built at Amgen under GxP. Veeva's agents run on Bedrock; UCB runs Azure — I'd test the Azure AI Foundry path first.

pre-call agent
next-best-action + reason visible
voice-to-CRM
human-confirmed
orchestration
journey triggers, one record
↓ recommend · approve · log ↓
CRM — SYSTEM OF RECORD

"Which agent first?" Voice capture — hours back, nothing customer-facing. "Veeva's wave one?" Pre-call, Voice, Free Text — plus PromoMats Quick Check and Content Agents, GA Dec 2025.

Q4 — ROI measurement~39 sec

"How do you measure ROI of agentic AI? Name the KPIs."

Baseline first, measurement plan in the charter, kill-or-scale gate at the pilot — no baseline, no pilot. Control group before anyone claims revenue. At Amgen: 40% drafting time saved, verified at a production milestone; that number earned Board approval. Small KPI set agreed with finance upfront, value tracking at every governance gate.

KPIWhy it leadsHow measured
Rep hours returnedFastest, cleanest valueWorkflow logs + shadowing, not self-report
NBA acceptance rateReps voting with clicksPaired with outcome sampling — metrics come in pairs
Channel engagement vs. baselineQuality signalPre/post per HCP segment
MLR cycle timeCompliance speed is valueTimestamp audit on AI-assisted content

UCB's CFO already attributes rising G&A to "accelerated digital transformation programs" — this program has a budget line and a watcher. Industry context: pharma AI spend heading $4B → $25.7B by 2030; the known failure mode is pilot purgatory, and the kill-or-scale gate is the exit.

"Would your 40% survive our CFO?" It survived a Board — time-per-document, same writers, same types, before/after. A stopwatch, not a survey.

Q5 — Six-month plan~37 sec

"Three vendors, a platform owner, skeptics — your plan?"

The first failure mode is decision rights, not technology — governance week one: Patient Impact owns priorities, DT AI owns the platform, I own delivery and the vendor scorecard, biweekly gate for spend decisions. Months 2–4: prove one workflow end to end, not three demos. Month six delivers a pilot in production with audited KPIs, a build-versus-buy decision, governance in writing.

M1 · audit + vendor scores + decision rights
M2–4 · one pilot proven end to end
M5–6 · harden, hand off to DT AI
Scorecardruns on our platform
respects CRM as system of record
pharma references
exit cost

Your questions back: "Where does SKAI sit against Veeva's agent roadmap — extend, build on Foundry, or coexist?" · "Which workflow hurts most today — pre-call prep, call notes, or MLR cycle?" · "What does month six look like in the CFO's language?"

"Why a contractor?" Concentrated senior delivery; capability stays in DT AI. "Skeptics?" On the review panel, their workflow as the pilot, their metric as the KPI. Proof: NewsRx — 8-week SOW scoped for four, delivered solo.

UCB stack

LayerTool
AgenticSKAI — Azure + Azure OpenAI + Foundry ("DT AI")
ProductivityMicrosoft 365 Copilot
Veeva liveMedical CRM · RIM · Safety Suite (June 2026)
Commercial CRMlegacy Salesforce-based → Vault CRM · Veeva AI agents, PromoMats, Approved Email, CLM · EOS Dec 31, 2029
DataAzure (off AWS) · Power BI · Synapse/Fabric
Not therePalantir, Databricks, Agentforce
SKAI — agents · Foundry · MLR gate · human approval · audit
↓ orchestrates around ↓
Veeva Vault CRM — system of record · pre-call & voice agents · PromoMats · Approved Email
↓ runs on ↓
Microsoft Azure — data estate · Power BI · Copilot

"Foundry hands-on?" Not hands-on — model-agnostic: selection, evaluation, orchestration. "Veeva regulatory?" Correct — Vault, RIM; commercial modules coming up to speed.

AWS → Azure

CategoryAWSAzure
ComputeEC2Virtual Machines
ServerlessLambdaAzure Functions
ContainersEKS / ECSAKS — K8s, Kubeflow
StorageS3Blob Storage
DatabaseRDS PostgreSQLAzure Database for PostgreSQL
NoSQLDynamoDBCosmos DB
WarehouseRedshiftSynapse / Fabric
ETLGlue / EMRData Factory
StreamingKinesisEvent Hubs
CategoryAWSAzure
MessagingSQS / SNSService Bus / Event Grid
MLSageMakerAzure Machine Learning
Managed LLMsBedrockAzure OpenAI / Foundry
IdentityIAMEntra ID + RBAC
NetworkVPCVirtual Network
MonitoringCloudWatchAzure Monitor
SecretsKMS / SecretsKey Vault
IaCCloudFormationARM / Bicep
BIQuickSightPower BI — UCB runs it
My AWS estate — RDS PostgreSQL · S3 · Redshift · SageMaker
↓ direct mapping (tables above) ↓
Azure landing zone — Azure PostgreSQL · Blob · Synapse/Fabric · Azure ML · Entra ID
↓ agents land on ↓
Azure OpenAI + Foundry — SKAI's layer